Codewhale is an open-source coding agent that runs in the terminal and can read repositories, edit files, execute commands, and inspect results while pursuing a task. Developers use it with hosted providers or local models, with controls for how much access it receives.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add Hmbown/Codewhale --skill cw-dogfoodgit clone --depth 1 https://github.com/Hmbown/CodewhaleWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/hmbown/codewhale/cw-dogfood)<a href="https://agentmods.dev/skills/hmbown/codewhale/cw-dogfood"><img src="https://agentmods.dev/badge/skills/hmbown/codewhale/cw-dogfood/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/hmbown/codewhale/cw-dogfood"><img src="https://agentmods.dev/badge/skills/hmbown/codewhale/cw-dogfood.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 106 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00052 | $0.01390 |
| Opus 5 | $0.00026 | $0.00695 |
| Sonnet 5 | $0.00010 | $0.00278 |
| Haiku 4.5 | $0.00005 | $0.00139 |
Grade A, and why
cw-dogfood scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 4d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 115 lines — stays where its author put it; the contents beside it link to each section on GitHub.
cw-dogfood
Green gates prove the code compiles and asserts. They do not prove the product works. Freezes, route contamination, focus theft, streaming cadence, and approval-flow regressions all live in the runtime, where no unit test looks. This stage puts the actual binary on your PATH and makes you use it.
Stage 4 of the loop: cw-orient → cw-slice → cw-gates → dogfood → cw-land → cw-handoff.
When to use
- The change is user-visible: TUI layout, motion, streaming, model or Fleet selection, approvals, commands, install paths.
- Before landing a release candidate, or before claiming a runtime behavior is fixed.
- Hunter asks to "install the build", "dogfood this", or "get this on my machine".
Workflow
-
Gate first. Run cw-gates to the rung the change deserves. Never install an ungated build.
-
Build stamped. Local builds are unstamped (
(dev)) since #5245, and the installer refuses an unstamped binary on purpose — the stamp is what proves the thing on your PATH is the thing you just built:CODEWHALE_BUILD_SHA=$(git rev-parse HEAD) \ cargo build --release --locked -p codewhale-cli -p codewhale-tui -
Install atomically. Use the script; do not hand-copy:
scripts/release/install-dogfood.sh # defaults to target/releaseIt refuses a dirty source tree (override deliberately with
CODEWHALE_ALLOW_DIRTY_DOGFOOD=1, and then say so wherever you report the install), verifies the binary embeds current HEAD, installscodewhaleandcodewinto~/.cargo/binand~/.local/bin(override withCODEWHALE_INSTALL_DIRS), re-signs ad-hoc on macOS, and verifies resolution from a fresh login shell.Never
cpover a running binary. On Apple Silicon that poisons the kernel's code-signature cache for the inode, and later execs hang until reboot. The installer does tmp-copy plus atomicmvfor exactly this reason.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 4d ago First seen · 115 lines · 52 tokens per session scan A 43d7c8f8d288
cw-dogfood is a skill published in the GitHub repository Hmbown/Codewhale (40,935 stars, last pushed today), licensed MIT. It adds 52 tokens to every session and 1,390 once invoked, about $0.0003 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-06.
Other skills, from other repositories
codew-release-qa-sweep
Use before claiming Codewhale release work is done: run the full gate sweep and list the manual QA targets.
verify
Exercise the real app/API/CLI and collect observable evidence; tests alone do not count as end-to-end verification.
webapp-testing
Start/reuse a local app, wait for readiness, inspect rendered state/console/network, act from observed selectors, and verify with evidence.
gh-find-prs
Survey open Codewhale PRs and triage each for mergeability and disposition against the real landing branch.
best-of-n
Generate a small set of independent candidate solutions in worktrees, judge them against one explicit rubric, and apply the winner only after PASS verification.
contributor-onboarding
Help a new contributor get productive on this checkout - inspect sync state against main, build, run the repository's exact verification gate, and produce a local what's-new digest. Never fetches, pulls, or modifies a dirty tree on its own. Explicit-only.